[BioC] QuasiSeq vs DSS

[Gordon K Smyth]

Dear Rich,

> Date: Tue, 12 Mar 2013 13:11:25 -0400
> From: Richard Friedman <friedman at cancercenter.columbia.edu>
> To: "Ryan C. Thompson" <rct at thompsonclan.org>
> Cc: Bioconductor mailing list <bioconductor at r-project.org>
> Subject: Re: [BioC] QuasiSeq vs DSS
>
> Dear Ryan,
>
> 	Thank you for your response.
> 3 questions:

> 1. If I had just a simple pairwise comparison is it known DSS or 
> QuasiSeq better?

> 2. I was unaware that an approximate implementation of QuasiSeq was 
> available in edgeR. If so, is it known how it compare to the ordinairy 
> EdgeR on the one hand and the full QuasiSeq on the other.

As Ryan has pointed out, the quasi-likelihood code in edgeR is not an 
approximation.  It is an independent implementation that leverages the 
additional capabilities of the limma and edgeR packages.  For example, the 
edgeR quasi-likelihood code on the devel repository has a robust empirical 
Bayes implementation that is very promising.

> 3. And I guess that the third question is for Gordon - Is using DSS and 
> QuasiSeq (or EdgeR) together desireable and if so, are there plans to 
> incorporate DSS into QuasiSeq (EdgeR).

No, one cannot combine DSS (or any other) shrinkage with quasi-likelihood 
shrinkage.  One cannot shrink values that have already been shrunk. There 
is no need need for this anyway.

> My note was planning ahead. I will still be in the microarray world for 
> a more few weeks before I return to learning RNASeq. I wanted to know 
> what the best practice is. If you (or anybody out there) develops a 
> script to meld the two methods, I am sure that it would be interesting 
> to the list.

I doubt that there is a single best method, although several methods are 
quite good.  RNA-seq data analysis is a very fast moving area and I don't 
think that it has settled down yet.  Most method authors probably feel 
that their own method is best.  However different people get different 
results from different simulation comparisons, so it would be wise to 
reserve judgement when new papers appear.  Moreover, what may be best for 
one dataset may not be best for another.

Best wishes
Gordon

> Best wishes,
> Rich
>
>
> On Mar 12, 2013, at 12:59 PM, Ryan C. Thompson wrote:
>
>> Dear Rich,

>> From what I can tell, it should be possible. The development version of 
>> DESeq2 implements the DSS "squeezing" method combined with edgeR's 
>> Cox-Reid dispersion estimation. You could use DESeq2 to estimate 
>> dispersions, and then copy those dispersion values into an edgeR 
>> DGEList object. Then you can use edgeR::glmQLFTest, which implements 
>> (approximately) the QuasiSeq method. I have not had time yet to 
>> investigate putting these packages together in this way, but it is 
>> something I plan to look at. I'm certain that the combination is 
>> technically possible, and I'm reasonably sure that the result would be 
>> statistically meaningful.

>> -Ryan Thompson

>> On Mar 12, 2013 7:06 AM, "Richard Friedman" <friedman at cancercenter.columbia.edu> wrote:

>> Dear List.
>>
>>         The papers on DSS (included in Bioconductor):
>>
>> Wu H, Wang C, Wu Z. A new shrinkage estimator for dispersion improves 
>> differential expression detection in RNA-seq data. Biostatistics. 2013 
>> Apr;14(2):232-43.
>>
>> and QuasiSeq (included in CRAN):
>>
>> Lund SP, Nettleton D, McCarthy DJ, Smyth GK. Detecting differential 
>> expression in RNA-sequence data using quasi-likelihood with shrunken 
>> dispersion estimates. Stat Appl Genet Mol Biol. 2012
>>
>> both give evidence of superior performance to edgeR (if I understand 
>> them correctly).
>>
>> Have the two methods been compared? Can the 2 methods been combined 
>> (with DSS estimating the dispersion used in the quasi-negative 
>> bionomial disribution used in QuasiSeq)?
>>
>> I would appreciate any insight with respect to what is the overall best 
>> method for differential expression in RNASeq available at present.
>>
>> Thanks and best wishes,
>> Rich
>>
>>
>> Richard A. Friedman, PhD
>> Associate Research Scientist,
>> Biomedical Informatics Shared Resource
>> Herbert Irving Comprehensive Cancer Center (HICCC)
>> Lecturer,
>> Department of Biomedical Informatics (DBMI)
>> Educational Coordinator,
>> Center for Computational Biology and Bioinformatics (C2B2)/
>> National Center for Multiscale Analysis of Genomic Networks (MAGNet)/
>> Columbia Initiative in Systems Biology
>> Room 824
>> Irving Cancer Research Center
>> Columbia University
>> 1130 St. Nicholas Ave
>> New York, NY 10032
>> (212)851-4765 (voice)
>> friedman at cancercenter.columbia.edu

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}